Explainable deep learning based techniques for ECG-Based heart disease classification: A systematic literature review...

Area of Science:

• Artificial Intelligence
• Cardiology
• Medical Informatics

Background:

• Deep learning (DL) models are increasingly used for classifying heart disease (HD) from electrocardiogram (ECG) data.
• Explainability in these DL models is crucial for clinical trust and understanding diagnostic reasoning.
• A systematic review is needed to understand the current landscape of explainable AI (XAI) applied to ECG-based HD classification.

Purpose of the Study:

• To systematically review and analyze the methodological choices in XAI-based DL architectures for ECG-based HD classification.
• To assess the impact of these choices on model interpretability and identify current challenges.
• To propose future research directions for improving explainability in this domain.

Main Methods:

• A systematic literature review (SLR) was conducted following Kitchenham and Charters guidelines.
• Academic articles published between January 2018 and September 2024 on XAI-based DL for ECG-based HD classification were analyzed.
• Insights into datasets, preprocessing methods, and XAI techniques were synthesized.

Main Results:

• Out of 6448 identified studies, 51 employed XAI-based DL architectures for ECG-based HD classification.
• The review identified 25 distinct datasets, 16 DL architectures, and 8 novel XAI techniques.
• Conventional XAI approaches like SHAP, Saliency Maps, Grad-Cam, and LIME were most frequently used.

Conclusions:

• Numerous XAI-based DL architectures exist for ECG-based HD classification, but significant limitations persist.
• Identified challenges include data standardization, inconsistent explainability, temporal dependency visualization, and lack of XAI benchmarking.
• Future research should focus on addressing these limitations to enhance the reliability and clinical utility of XAI in ECG analysis.